Introduction

Our purpose is to automatically extract tree crowns from high resolution aerial images (cf Fig. (1)) in order to assess some parameters such as the number of trees, their diameter and the stem density.

Figure 1: Left : some data provided by IFN, in Saône et Loire (France). Right : zoom on a plantation of poplars (left) and some oaks (right). © IFN

Our approach consists in modeling the forestry image as a realization of a marked point process of ellipses, where the points represent the positions of the trees and the marks their geometric shape. The goal is to find the configuration of an unknown number of objects which maximizes the density of the process. To that prospect we use some Markov Chain Monte Carlo (MCMC) algorithms and Simulated Annealing (SA) for simulation and optimization.